TouchReal: an online neuromorphic tactile framework for biomimetic afferent spike generation based on TouchSim

Robotic tactile sensors usually provide continuous, sensor-specific signals such as force, pressure, or vibration. Although such signals are useful for many perception tasks, they are not directly compatible with afferent-level neural representations used in biological touch or spike-based processing systems. Here, we present TouchReal, an online neuromorphic tactile framework that converts signals from a low-cost multimodal tactile sensor into biomimetic SA1-, RA-, and PC-like spike trains inspired by the neural simulator package TouchSim. The hardware combines a capacitive SingleTact force sensor for static contact information with a PVDF piezoelectric film for dynamic vibration-sensitive responses. Raw sensor signals are processed through a pre-processing pipeline and stateful leaky integrate-and-fire (LIF) afferent models. The generated spike trains reproduce several classical temporal, frequency-dependent, and receptive-field-like response properties of cutaneous afferents at an average absolute latency of 130.02 ms with a 32 ms update interval. Functional verification further shows that SA1 and RA spikes support force regression, while PC spikes enable texture classification with around 90% accuracy using temporal encoding features. These results suggest that TouchReal provides an interpretable afferent-level tactile representation for biomimetic robotic sensing, opening up possibilities for neuroprosthetic feedback and robotic models of human perception.